Pub Date : 2022-06-06DOI: 10.1109/aiiot54504.2022.9817165
B. Gowrienanthan, N. Kiruthihan, K. Rathnayake, S. Kumarawadu, V. Logeeshan
Non-Intrusive Load Monitoring (NILM) is the process of monitoring the power consumption of individual appliances by disaggregating the aggregate power consumption data from a single sensor, which is usually the main meter. The increase in adoption of smart meters facilitates large scale NILM. Appliance-level load monitoring could provide utilities and users with useful information which could lead to significant energy savings as well as better demand-side management. In this paper, we propose a low-cost method for ensembling deep neural network models trained for the task of load disaggregation, which does not require the training of multiple different models. Additionally, we analyze the output characteristics of the resultant ensembled model in relation to the outputs of its component models. The UK-DALE dataset is used for training the models and evaluating the effectiveness of our ensembling technique. The results show that the proposed technique provides a considerable improvement in load disaggregation performance.
{"title":"Low-Cost Ensembling for Deep Neural Network based Non-Intrusive Load Monitoring","authors":"B. Gowrienanthan, N. Kiruthihan, K. Rathnayake, S. Kumarawadu, V. Logeeshan","doi":"10.1109/aiiot54504.2022.9817165","DOIUrl":"https://doi.org/10.1109/aiiot54504.2022.9817165","url":null,"abstract":"Non-Intrusive Load Monitoring (NILM) is the process of monitoring the power consumption of individual appliances by disaggregating the aggregate power consumption data from a single sensor, which is usually the main meter. The increase in adoption of smart meters facilitates large scale NILM. Appliance-level load monitoring could provide utilities and users with useful information which could lead to significant energy savings as well as better demand-side management. In this paper, we propose a low-cost method for ensembling deep neural network models trained for the task of load disaggregation, which does not require the training of multiple different models. Additionally, we analyze the output characteristics of the resultant ensembled model in relation to the outputs of its component models. The UK-DALE dataset is used for training the models and evaluating the effectiveness of our ensembling technique. The results show that the proposed technique provides a considerable improvement in load disaggregation performance.","PeriodicalId":409264,"journal":{"name":"2022 IEEE World AI IoT Congress (AIIoT)","volume":"186 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121423018","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-06DOI: 10.1109/aiiot54504.2022.9817338
Md Rakibul Ahasan, Mirza Sanita Haque, Mohammad Rubbyat Akram, Mohammed Fahim Momen, Md. Golam Rabiul Alam
A 4G network stands for a fourth-generation mobile network that enables 4G capable mobile phones to connect with the internet faster than ever. It is possible because of faster authentication between mobile phone and network entity. The network entities are sophisticated and require constant monitoring in terms of fault management and performance management. However, the fault is very rare in that network nodes, but a deviation of performance is normal. This deviation is known as an anomaly, and machine learning is useful for detecting an anomaly. In this paper, deep neural network autoencoder-based anomaly detection is discussed over 4G network performance data. An autoencoder can mimic an output from its input and provide superior performance when the data properties are similar. Further elaboration in this paper is how different properties of autoencoder hidden layer count, variable threshold measurement etc influence the anomaly detection outcome of 4G network performance data. At last, an autoencoder configuration is recommended for anomaly detection of 4G network performance data.
{"title":"Deep Learning Autoencoder based Anomaly Detection Model on 4G Network Performance Data","authors":"Md Rakibul Ahasan, Mirza Sanita Haque, Mohammad Rubbyat Akram, Mohammed Fahim Momen, Md. Golam Rabiul Alam","doi":"10.1109/aiiot54504.2022.9817338","DOIUrl":"https://doi.org/10.1109/aiiot54504.2022.9817338","url":null,"abstract":"A 4G network stands for a fourth-generation mobile network that enables 4G capable mobile phones to connect with the internet faster than ever. It is possible because of faster authentication between mobile phone and network entity. The network entities are sophisticated and require constant monitoring in terms of fault management and performance management. However, the fault is very rare in that network nodes, but a deviation of performance is normal. This deviation is known as an anomaly, and machine learning is useful for detecting an anomaly. In this paper, deep neural network autoencoder-based anomaly detection is discussed over 4G network performance data. An autoencoder can mimic an output from its input and provide superior performance when the data properties are similar. Further elaboration in this paper is how different properties of autoencoder hidden layer count, variable threshold measurement etc influence the anomaly detection outcome of 4G network performance data. At last, an autoencoder configuration is recommended for anomaly detection of 4G network performance data.","PeriodicalId":409264,"journal":{"name":"2022 IEEE World AI IoT Congress (AIIoT)","volume":"144 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116414925","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-06DOI: 10.1109/aiiot54504.2022.9817249
David W. Mayo, H. Elgazzar
Cryptocurrency prices are highly variable. Predicting changes in cryptocurrency price is a hugely important topic to investors and researchers, with much existing research on demand-side factors. The goal of this research project is to design and implement machine learning models to predict future cryptocurrency price change direction based primarily on supply-side factors. Different unsupervised machine learning techniques are used to build the predictive models. These techniques include K Nearest Neighbors (KNN), Artificial Neural Networks (ANN), Support Vector Machines (SVM), Naive Bayesian Classifier, and Random Forest Classifier. A dataset of 10 daily supply-side metrics for three prominent cryptocurrencies (Bitcoin, Ethereum, and Litecoin) at four different time horizons (ranging from one day to 30 days) are used to build and test the machine learning models. The outputs of these models indicate the predicted direction of the price movement over the time horizon (i.e., whether the price would go up or down), not the magnitude of the movement. Experimental results show that predictions were very unreliable for the shorter time spans but very reliable for the longest time spans. The Artificial Neural Network and Random Forest classifiers consistently outperformed the other techniques and achieved a prediction accuracy of over 90% in most models and over 95% in the best models. Experimental results show also that there is no significant difference in predictability between the three prominent cryptocurrencies.
{"title":"Predicting Cryptocurrency Price Change Direction from Supply-Side Factors via Machine Learning Methods","authors":"David W. Mayo, H. Elgazzar","doi":"10.1109/aiiot54504.2022.9817249","DOIUrl":"https://doi.org/10.1109/aiiot54504.2022.9817249","url":null,"abstract":"Cryptocurrency prices are highly variable. Predicting changes in cryptocurrency price is a hugely important topic to investors and researchers, with much existing research on demand-side factors. The goal of this research project is to design and implement machine learning models to predict future cryptocurrency price change direction based primarily on supply-side factors. Different unsupervised machine learning techniques are used to build the predictive models. These techniques include K Nearest Neighbors (KNN), Artificial Neural Networks (ANN), Support Vector Machines (SVM), Naive Bayesian Classifier, and Random Forest Classifier. A dataset of 10 daily supply-side metrics for three prominent cryptocurrencies (Bitcoin, Ethereum, and Litecoin) at four different time horizons (ranging from one day to 30 days) are used to build and test the machine learning models. The outputs of these models indicate the predicted direction of the price movement over the time horizon (i.e., whether the price would go up or down), not the magnitude of the movement. Experimental results show that predictions were very unreliable for the shorter time spans but very reliable for the longest time spans. The Artificial Neural Network and Random Forest classifiers consistently outperformed the other techniques and achieved a prediction accuracy of over 90% in most models and over 95% in the best models. Experimental results show also that there is no significant difference in predictability between the three prominent cryptocurrencies.","PeriodicalId":409264,"journal":{"name":"2022 IEEE World AI IoT Congress (AIIoT)","volume":"73 6 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131012837","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-06DOI: 10.1109/aiiot54504.2022.9817145
Kabir Hossain, Tonmoy Ghosh, E. Sazonov
This paper focuses on development of server-based infrastructure for real-time analysis of wearable signals. In this work, we have implemented a python flask-based API (Application Programming Interface) to receive sensor and image data from various platforms (e.g., mobile, computer), and created a data storage (MariaDB database and file server) to store data. A load balancer, Nginx, that redirects traffic into different ports was configured for low latency. Additionally, we developed a food intake detection method based on machine learning (ML). We have investigated ten different ML models to find an accurate and fast model. To test the server infrastructure, we conducted a functionality test to verify each component of the server. We also investigated how a number of APIs influence the performance of the server in terms of latency. To verify the server, we performed a computer simulation where a python script was used to deliver signals and images continuously to the server. We sent a total of five hundred images and sensor signals to the server from two different processes simultaneously. We achieved an average latency of 260ms and 110ms for signal and image packets, respectively. The average latency decreased by 26.92% and 15.38% when we use two API ports. For food intake detections, data were collected from 17 free-living (9 males, 6 females, and 2 adolescents) volunteers. Thereafter these data were evaluated by ten different ML classifiers, e.g., Adaboost (AB), Random Forest (RF), Gradient Boosting (GB) and Histogram Gradient Boosting (HGB). The experiments were performed by 5-fold validations, where 80% of subjects were used for training the remaining 20% for testing. The RF model provided the best result with average accuracy, precision, recall and F1-score of 0.99, 0.97, 0.97 and 0.98, respectively. Results indicate that our implemented server architecture was able to receive signals in real-time and detect food intake with high accuracy.
{"title":"Development of Cloud-based Infrastructure for Real Time Analysis of Wearable Sensor Signal","authors":"Kabir Hossain, Tonmoy Ghosh, E. Sazonov","doi":"10.1109/aiiot54504.2022.9817145","DOIUrl":"https://doi.org/10.1109/aiiot54504.2022.9817145","url":null,"abstract":"This paper focuses on development of server-based infrastructure for real-time analysis of wearable signals. In this work, we have implemented a python flask-based API (Application Programming Interface) to receive sensor and image data from various platforms (e.g., mobile, computer), and created a data storage (MariaDB database and file server) to store data. A load balancer, Nginx, that redirects traffic into different ports was configured for low latency. Additionally, we developed a food intake detection method based on machine learning (ML). We have investigated ten different ML models to find an accurate and fast model. To test the server infrastructure, we conducted a functionality test to verify each component of the server. We also investigated how a number of APIs influence the performance of the server in terms of latency. To verify the server, we performed a computer simulation where a python script was used to deliver signals and images continuously to the server. We sent a total of five hundred images and sensor signals to the server from two different processes simultaneously. We achieved an average latency of 260ms and 110ms for signal and image packets, respectively. The average latency decreased by 26.92% and 15.38% when we use two API ports. For food intake detections, data were collected from 17 free-living (9 males, 6 females, and 2 adolescents) volunteers. Thereafter these data were evaluated by ten different ML classifiers, e.g., Adaboost (AB), Random Forest (RF), Gradient Boosting (GB) and Histogram Gradient Boosting (HGB). The experiments were performed by 5-fold validations, where 80% of subjects were used for training the remaining 20% for testing. The RF model provided the best result with average accuracy, precision, recall and F1-score of 0.99, 0.97, 0.97 and 0.98, respectively. Results indicate that our implemented server architecture was able to receive signals in real-time and detect food intake with high accuracy.","PeriodicalId":409264,"journal":{"name":"2022 IEEE World AI IoT Congress (AIIoT)","volume":"6 11","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132545542","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-06DOI: 10.1109/aiiot54504.2022.9817271
Md Mahadi Hassan Sohan, Mohammad Monirujjaman Khan, Ipseeta Nanda, Rajesh Dey
Online reviews play a crucial role in determining whether a product will be sold on e-commerce websites or applications. Because so many people rely on internet evaluations, unethical actors may fabricate reviews in order to artificially boost or devalue items and services. To detect false product reviews, this research provides a semi-supervised machine learning approach. Furthermore, feature engineering techniques are used in this work to extract diverse reviewer behaviors. This study examines the outcomes of numerous experiments on a real food review dataset of restaurant reviews with attributes collected from user behavior. In terms off-score, the results indicate that Random Forest surpasses another classifier, with the best f-score of 98 %. In addition, the data reveals that taking into account the reviewers' behavioral characteristics raises the f-score and the final accuracy has come out 97.7%. In the current technique, not all reviewers' behavioral characteristics have been considered. Other low-level features such as frequent time or date dependency, the reviewer's timing for giving a review, and how common it is to deliver favorable or poor reviews will be added further in order to improve the efficacy of the offered fake review detecting algorithm.
{"title":"Fake Product Review Detection Using Machine Learning","authors":"Md Mahadi Hassan Sohan, Mohammad Monirujjaman Khan, Ipseeta Nanda, Rajesh Dey","doi":"10.1109/aiiot54504.2022.9817271","DOIUrl":"https://doi.org/10.1109/aiiot54504.2022.9817271","url":null,"abstract":"Online reviews play a crucial role in determining whether a product will be sold on e-commerce websites or applications. Because so many people rely on internet evaluations, unethical actors may fabricate reviews in order to artificially boost or devalue items and services. To detect false product reviews, this research provides a semi-supervised machine learning approach. Furthermore, feature engineering techniques are used in this work to extract diverse reviewer behaviors. This study examines the outcomes of numerous experiments on a real food review dataset of restaurant reviews with attributes collected from user behavior. In terms off-score, the results indicate that Random Forest surpasses another classifier, with the best f-score of 98 %. In addition, the data reveals that taking into account the reviewers' behavioral characteristics raises the f-score and the final accuracy has come out 97.7%. In the current technique, not all reviewers' behavioral characteristics have been considered. Other low-level features such as frequent time or date dependency, the reviewer's timing for giving a review, and how common it is to deliver favorable or poor reviews will be added further in order to improve the efficacy of the offered fake review detecting algorithm.","PeriodicalId":409264,"journal":{"name":"2022 IEEE World AI IoT Congress (AIIoT)","volume":"58 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134545628","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-06DOI: 10.1109/aiiot54504.2022.9817352
Argeen Blanco, Lance Victor Del Rosario, Ken Ichiro Jose, Melchizedek I. Alipio
According to the World Bank, one out of five Filipinos do not get water from formal sources. Only 77% of the rural population and 90% of those in urban areas have access to an improved water source and only 44% have direct house connections. Surveillance of water quality is mandatory thus many research studies have been presented to different communities that showed effective results. In rural areas, there is already a classification model for water potability using traditional machine learning techniques. However, there currently no deep learning-based model for water potability classification. Thus, this work aims to create a deep learning water potability classification model for rural water sources in the Philippines. It starts from importing the water potability dataset of water monitoring sources from rural areas then pre-processing of the data, evaluation of the performance of the learning models through accuracy, precision, recall and f-measure metrics. Out of all the three, MLP had provided the greatest accuracy of 99.80%. LSTM performed better in accuracy and recall in comparison to GRU, but GRU had provided better precision than LSTM. LSTM has been considered to greatly classify the most common classifications in the dataset, while GRU has been observed to accurately classify the infrequent classifications in the dataset.
{"title":"Deep Learning Models for Water Potability Classification in Rural Areas in the Philippines","authors":"Argeen Blanco, Lance Victor Del Rosario, Ken Ichiro Jose, Melchizedek I. Alipio","doi":"10.1109/aiiot54504.2022.9817352","DOIUrl":"https://doi.org/10.1109/aiiot54504.2022.9817352","url":null,"abstract":"According to the World Bank, one out of five Filipinos do not get water from formal sources. Only 77% of the rural population and 90% of those in urban areas have access to an improved water source and only 44% have direct house connections. Surveillance of water quality is mandatory thus many research studies have been presented to different communities that showed effective results. In rural areas, there is already a classification model for water potability using traditional machine learning techniques. However, there currently no deep learning-based model for water potability classification. Thus, this work aims to create a deep learning water potability classification model for rural water sources in the Philippines. It starts from importing the water potability dataset of water monitoring sources from rural areas then pre-processing of the data, evaluation of the performance of the learning models through accuracy, precision, recall and f-measure metrics. Out of all the three, MLP had provided the greatest accuracy of 99.80%. LSTM performed better in accuracy and recall in comparison to GRU, but GRU had provided better precision than LSTM. LSTM has been considered to greatly classify the most common classifications in the dataset, while GRU has been observed to accurately classify the infrequent classifications in the dataset.","PeriodicalId":409264,"journal":{"name":"2022 IEEE World AI IoT Congress (AIIoT)","volume":"327 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134057038","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-06DOI: 10.1109/aiiot54504.2022.9817355
M. Uddin, Raphael Pamie-George, Daron Wilkins, Andres Sousa-Poza, M. Canan, Samuel F. Kovacic, Jiang Li
Generative adversarial networks (GANs) have become very popular in recent years. GANs have proved to be successful in different computer vision tasks including image-translation, image super-resolution etc. In this paper, we have used GAN models for ship deck segmentation. We have used 2D scanned raster images of ship decks provided by US Navy Military Sealift Command (MSC) to extract necessary information including ship walls, objects etc. Our segmentation results will be helpful to get vector and 3D image of a ship that can be later used for maintenance of the ship. We applied the trained models to engineering documents provided by MSC and obtained very promising results, demonstrating that GANs can be potentially good candidates for this research area.
{"title":"Ship Deck Segmentation In Engineering Document Using Generative Adversarial Networks","authors":"M. Uddin, Raphael Pamie-George, Daron Wilkins, Andres Sousa-Poza, M. Canan, Samuel F. Kovacic, Jiang Li","doi":"10.1109/aiiot54504.2022.9817355","DOIUrl":"https://doi.org/10.1109/aiiot54504.2022.9817355","url":null,"abstract":"Generative adversarial networks (GANs) have become very popular in recent years. GANs have proved to be successful in different computer vision tasks including image-translation, image super-resolution etc. In this paper, we have used GAN models for ship deck segmentation. We have used 2D scanned raster images of ship decks provided by US Navy Military Sealift Command (MSC) to extract necessary information including ship walls, objects etc. Our segmentation results will be helpful to get vector and 3D image of a ship that can be later used for maintenance of the ship. We applied the trained models to engineering documents provided by MSC and obtained very promising results, demonstrating that GANs can be potentially good candidates for this research area.","PeriodicalId":409264,"journal":{"name":"2022 IEEE World AI IoT Congress (AIIoT)","volume":"6 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123798513","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-06DOI: 10.1109/aiiot54504.2022.9817282
I.B.C. Irugalbandara, Adil Naseem, Melaka Perera, V. Logeeshan
People can follow a more comfortable lifestyle, thanks to the improvements in the Internet of Things (IoT) technologies, making it easier to operate and monitor their electrical and electronic products at home. Even the elderly can utilize home automation systems with a simple voice command. Disabled people are also getting the most benefits from these systems. Many home automation systems now rely on cloud-based services when it comes to features like voice assistants. Because these services transmit personal data to cloud services via the internet, home automation systems require a stable internet connection and a secure environment free of cyberattacks. Additionally, users of these systems cannot make full use of them because the internet quality index is generally low in developing nations. This study presents an offline home automation system to address these difficulties. Without the internet or cloud services, the proposed home automation system can perform its essential functions. It also offers additional features like power tracking and optimization in linked devices while ensuring protection against foreign assaults and giving quick responses.
{"title":"HomeIO: Offline Smart Home Automation System with Automatic Speech Recognition and Household Power Usage Tracking","authors":"I.B.C. Irugalbandara, Adil Naseem, Melaka Perera, V. Logeeshan","doi":"10.1109/aiiot54504.2022.9817282","DOIUrl":"https://doi.org/10.1109/aiiot54504.2022.9817282","url":null,"abstract":"People can follow a more comfortable lifestyle, thanks to the improvements in the Internet of Things (IoT) technologies, making it easier to operate and monitor their electrical and electronic products at home. Even the elderly can utilize home automation systems with a simple voice command. Disabled people are also getting the most benefits from these systems. Many home automation systems now rely on cloud-based services when it comes to features like voice assistants. Because these services transmit personal data to cloud services via the internet, home automation systems require a stable internet connection and a secure environment free of cyberattacks. Additionally, users of these systems cannot make full use of them because the internet quality index is generally low in developing nations. This study presents an offline home automation system to address these difficulties. Without the internet or cloud services, the proposed home automation system can perform its essential functions. It also offers additional features like power tracking and optimization in linked devices while ensuring protection against foreign assaults and giving quick responses.","PeriodicalId":409264,"journal":{"name":"2022 IEEE World AI IoT Congress (AIIoT)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130624063","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-06DOI: 10.1109/aiiot54504.2022.9817237
Abdullah Al-Monsur, Md Rizwanul Kabir, Abrar Mohammad Ar-Rafi, M. M. Nishat, Fahim Faisal
Covid-19 is still running rampant around the globe. With the recent emergence of rapidly spreading variants, the necessity for testing becomes ever more acute. In this study, firstly, a deep learning based framework is proposed to conduct both a binary and multi-class classification of chest X-ray images to detect Covid-19 in order to meet the demands of swift, accurate testing worldwide. It is carried out using Convolutional Neural Networks to comprehensively examine the Covid-19 Chest X-ray dataset in conjunction with X-ray images of lungs with pneumonia. The architecture developed for the classification process is termed as CovidNet and its performance is compared with the existing Vgg16 architecture. Secondly, in order to obtain an enhanced performance, the proposed CovidNet is coupled with the Vgg16 architecture by means of ensembling to produce the Covid-EnsembleNet model. In the binary classification process, the developed CovidNet architecture results in a test accuracy of 99.12% while the Vgg16 architecture performs with a 99.34% accuracy. The Covid-EnsembleNet yields an accuracy of 99.56% in this process thereby bolstering the proposed model. Afterwards, in the multi-class classification process the CovidNet achieves a test accuracy of 94.96 % with the Vgg16 achieving a test accuracy of 96.75%. The proposed ensemble model Covid-EnsembleNet yields a test accuracy 97.56 %, thereby, outperforming both the CovidNet and existing Vgg16 architecture in both types of classification.
{"title":"Covid-EnsembleNet: An Ensemble Based Approach for Detecting Covid-19 by utilising Chest X-ray Images","authors":"Abdullah Al-Monsur, Md Rizwanul Kabir, Abrar Mohammad Ar-Rafi, M. M. Nishat, Fahim Faisal","doi":"10.1109/aiiot54504.2022.9817237","DOIUrl":"https://doi.org/10.1109/aiiot54504.2022.9817237","url":null,"abstract":"Covid-19 is still running rampant around the globe. With the recent emergence of rapidly spreading variants, the necessity for testing becomes ever more acute. In this study, firstly, a deep learning based framework is proposed to conduct both a binary and multi-class classification of chest X-ray images to detect Covid-19 in order to meet the demands of swift, accurate testing worldwide. It is carried out using Convolutional Neural Networks to comprehensively examine the Covid-19 Chest X-ray dataset in conjunction with X-ray images of lungs with pneumonia. The architecture developed for the classification process is termed as CovidNet and its performance is compared with the existing Vgg16 architecture. Secondly, in order to obtain an enhanced performance, the proposed CovidNet is coupled with the Vgg16 architecture by means of ensembling to produce the Covid-EnsembleNet model. In the binary classification process, the developed CovidNet architecture results in a test accuracy of 99.12% while the Vgg16 architecture performs with a 99.34% accuracy. The Covid-EnsembleNet yields an accuracy of 99.56% in this process thereby bolstering the proposed model. Afterwards, in the multi-class classification process the CovidNet achieves a test accuracy of 94.96 % with the Vgg16 achieving a test accuracy of 96.75%. The proposed ensemble model Covid-EnsembleNet yields a test accuracy 97.56 %, thereby, outperforming both the CovidNet and existing Vgg16 architecture in both types of classification.","PeriodicalId":409264,"journal":{"name":"2022 IEEE World AI IoT Congress (AIIoT)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130680407","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-06DOI: 10.1109/aiiot54504.2022.9817324
Saroj Mishra, H. Reza
At the present, the use of face masks is growing day by day and it is mandated in most places across the world. People are encouraged to cover their faces when in public areas to avoid the spread of infection which can minimize the transmission of Covid-19 by 65 percent (according to the public health officials). So, it is important to detect people not wearing face masks. Additionally, face recognition has been applied to a wide area for security verification purposes since its performance, accuracy, and reliability [15] are better than any other traditional techniques like fingerprints, passwords, PINs, and so on. In recent years, facial recognition is becoming a challenging task because of various occlusions or masks like the existence of sunglasses, scarves, hats, and the use of make-up or disguise ingredients. So, the face recognition accuracy rate is affected by these types of masks. Moreover, the use of face masks has made conventional facial recognition technology ineffective in many scenarios, such as face authentication, security check, tracking school, and unlocking phones and laptops. As a result, we proposed a solution, Masked Facial Recognition (MFR) which can identify masked and unmasked people so individuals wearing a face mask do not need to take it out to authenticate themselves. We used the Deep Learning model, Inception ResNet V1 to train our model. The CASIA dataset [17] is applied for training images and the LFW (Labeled Faces in the Wild) dataset [18] is used for model evaluation purposes. The masked datasets are created using a Computer Vision-based approach (Dlib), We received an accuracy of over 96 percent for our three different trained models. As a result, the purposed work could be utilized effortlessly for both masked and unmasked face recognition and detection systems that are designed for safety and security verification purposes without any challenges.
目前，口罩的使用日益增加，在世界上大多数地方都是强制性的。鼓励人们在公共场所遮住脸，以避免感染的传播，这可以将Covid-19的传播减少65%(根据公共卫生官员的说法)。因此，检测不戴口罩的人很重要。此外，人脸识别由于其性能、准确性和可靠性[15]优于指纹、密码、pin等任何传统技术，已被广泛应用于安全验证目的。近年来，面部识别正成为一项具有挑战性的任务，因为存在各种遮挡或面具，如太阳镜、围巾、帽子，以及使用化妆品或伪装成分。因此，人脸识别的准确率受到这些类型面具的影响。此外，口罩的使用使得传统的面部识别技术在许多场景中无效，例如面部认证，安全检查，跟踪学校，解锁手机和笔记本电脑。因此，我们提出了一种解决方案，即蒙面人脸识别(MFR)，它可以识别蒙面和未蒙面的人，这样戴着口罩的人就不需要拿出来进行身份验证。我们使用深度学习模型Inception ResNet V1来训练我们的模型。CASIA数据集[17]用于训练图像，LFW (Labeled Faces in The Wild)数据集[18]用于模型评估。掩蔽数据集是使用基于计算机视觉的方法(Dlib)创建的，我们对三种不同的训练模型获得了超过96%的准确率。因此，可以毫不费力地将目标工作用于为安全和保安核查目的而设计的蒙面和未蒙面人脸识别和检测系统，而不会遇到任何挑战。
{"title":"A Face Recognition Method Using Deep Learning to Identify Mask and Unmask Objects","authors":"Saroj Mishra, H. Reza","doi":"10.1109/aiiot54504.2022.9817324","DOIUrl":"https://doi.org/10.1109/aiiot54504.2022.9817324","url":null,"abstract":"At the present, the use of face masks is growing day by day and it is mandated in most places across the world. People are encouraged to cover their faces when in public areas to avoid the spread of infection which can minimize the transmission of Covid-19 by 65 percent (according to the public health officials). So, it is important to detect people not wearing face masks. Additionally, face recognition has been applied to a wide area for security verification purposes since its performance, accuracy, and reliability [15] are better than any other traditional techniques like fingerprints, passwords, PINs, and so on. In recent years, facial recognition is becoming a challenging task because of various occlusions or masks like the existence of sunglasses, scarves, hats, and the use of make-up or disguise ingredients. So, the face recognition accuracy rate is affected by these types of masks. Moreover, the use of face masks has made conventional facial recognition technology ineffective in many scenarios, such as face authentication, security check, tracking school, and unlocking phones and laptops. As a result, we proposed a solution, Masked Facial Recognition (MFR) which can identify masked and unmasked people so individuals wearing a face mask do not need to take it out to authenticate themselves. We used the Deep Learning model, Inception ResNet V1 to train our model. The CASIA dataset [17] is applied for training images and the LFW (Labeled Faces in the Wild) dataset [18] is used for model evaluation purposes. The masked datasets are created using a Computer Vision-based approach (Dlib), We received an accuracy of over 96 percent for our three different trained models. As a result, the purposed work could be utilized effortlessly for both masked and unmasked face recognition and detection systems that are designed for safety and security verification purposes without any challenges.","PeriodicalId":409264,"journal":{"name":"2022 IEEE World AI IoT Congress (AIIoT)","volume":"55 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121976451","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：